How to much visual classification using nonparametric intensities? A behavioral answer

Investigates a fully nonparametric supervised classification scheme underlined by a nonhomogeneous poisson point process for the distribution of points in the radiometric space. Uniform kernel estimates of the intensity functions are investigated. In particular, the choice of the kernel bandwidth is discussed. The a priori probability for a point to belong to one of the populations is estimated via an EM (expectation maximization) algorithm in which the intensity (or density) function estimates are kept constant. Finally, the authors propose a contextual ICM (Iterated Conditional mode) classifier underlined by a Markov random field model for the distribution of points on the 2-geometrical lattice and including computation of the density function estimates, estimation of the kernel bandwidth and computation of the a priori probabilities